Revert "macos: use posix_spawn instead of fork" by cjihrig · Pull Request #3107 · libuv/libuv

cjihrig · 2021-02-12T17:16:51Z

This reverts commit 39968db.

This commit caused a number of failures in the Node.js test suite
while attempting to release libuv 1.41.0.

Node.js integration CI: https://ci.nodejs.org/view/libuv/job/libuv-in-node/175/. The previously observed macOS failures seem to be fixed by this revert.

bnoordhuis

Regrettable but it is what it is. Let's revert and revisit.

vtjnash

Unfortunate that jpcanepa is on vacation now, but I expect he'll dig into the findings soon after he's back. And better to get the release done. Sorry it delayed it so much without effect!

cjihrig · 2021-02-12T22:01:48Z

A rare green libuv CI run 😄 : https://ci.nodejs.org/view/libuv/job/libuv-test-commit/2125/

This reverts commit 39968db. This commit caused a number of failures in the Node.js test suite while attempting to release libuv 1.41.0. Refs: libuv#3086 Refs: libuv#3064 PR-URL: libuv#3107 Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl> Reviewed-By: Jameson Nash <vtjnash@gmail.com> Reviewed-By: Santiago Gimeno <santiago.gimeno@gmail.com>

vtjnash · 2021-02-12T22:10:09Z

Some day I'm hoping we could install rr on these Jenkins workers, and then any linux failure would be trivial to reproduce, diagnose, and fix. We've been using it on our buildbots for JuliaLang for the past couple of months, and it has helped enormously at driving down our rare failure rate. The author also has some advance integration tooling for working with javascript, which might be interesting to the nodejs folks too. I don't really know who to contact for installing and setting this up however.

cjihrig · 2021-02-12T22:12:15Z

If you open an issue in https://github.com/nodejs/build they should be able to help (assuming they think it's a good idea). cc: @richardlau

richardlau · 2021-02-12T23:42:52Z

Assuming that it is https://github.com/rr-debugger/rr that looks to be Linux only and the issue with this one was a broken macOS CI?

Feel free to open an issue in build if you feel rr would be useful -- it would be helpful to also explain how it would be used (e.g. would we need to upload new artifacts (the recordings) and how much space those are likely to need) bearing in mind that most users who are able to start jobs in Jenkins don't have permissions to log into the machines themselves (temporary access can be granted on request).

If it's Linux another option would be GitHub actions.

deepak1556 · 2021-02-18T19:30:27Z

@cjihrig just catching up on this, can you point out which tests in Node.js suite were failing ?

Electron backported the posix_spawn change to its stable release lines last week, would be good to know why it should be reverted. Thanks!

cjihrig · 2021-02-18T19:34:38Z

Here are the failures: https://ci.nodejs.org/job/node-test-commit-osx/38851/. Please note that the CI results are not persisted permanently, so I don't know how much longer that link will be valid.

For posterity, the failures are:

test.parallel/test-child-process-cwd
test.parallel/test-child-process-exec-error
test.parallel/test-child-process-promisified
test.parallel/test-child-process-spawn-error
test.parallel/test-child-process-spawnsync-validation-errors
test.parallel/test-heapsnapshot-near-heap-limit-worker
test.parallel/test-child-process-exec-error
test.parallel/test-child-process-promisified
test.parallel/test-child-process-spawn-error

test-heapsnapshot-near-heap-limit-worker may not be related though.

deepak1556 · 2021-02-20T00:47:40Z

Looked into the failing tests there were 3 different types, summarizing what I have found so far.

The following tests were failing when they hit a runtime errors but libuv was closing the pipe that didn't allow for further processing in https://github.com/nodejs/node/blob/master/lib/internal/child_process.js#L444-L456

test.parallel/test-child-process-cwd
test.parallel/test-child-process-exec-error
test.parallel/test-child-process-promisified
test.parallel/test-child-process-spawn-error

I tried to match the previous behavior for these cases and it seems to fix the above tests

diff --git a/deps/uv/src/unix/process.c b/deps/uv/src/unix/process.c
index b4e44b049f..abc83e24c9 100644
--- a/deps/uv/src/unix/process.c
+++ b/deps/uv/src/unix/process.c
@@ -970,7 +970,12 @@ int uv_spawn(uv_loop_t* loop,
 
   /* Spawn the child */
   err = uv__spawn_and_init_child(options, stdio_count, pipes, signal_pipe[1], &pid);
-  if (err != 0) {
+  if (err != 0 &&
+      !(err == UV_EACCES ||
+       err == UV_EAGAIN ||
+       err == UV_EMFILE ||
+       err == UV_ENFILE ||
+       err == UV_ENOENT)) {
     uv_rwlock_wrunlock(&loop->cloexec_lock);
     uv__close(signal_pipe[0]);
     uv__close(signal_pipe[1]);
@@ -982,7 +987,7 @@ int uv_spawn(uv_loop_t* loop,
   uv__close(signal_pipe[1]);
 
   process->status = 0;
-  exec_errorno = 0;
+  exec_errorno = err;
   do
     r = read(signal_pipe[0], &exec_errorno, sizeof(exec_errorno));
   while (r == -1 && errno == EINTR);

test.parallel/test-child-process-stdio was failing with EINTR at the posix_spawn call, I don't know what causes this but adding a retry logic fixes this case.

diff --git a/deps/uv/src/unix/process.c b/deps/uv/src/unix/process.c
index b4e44b049f..241272a806 100644
--- a/deps/uv/src/unix/process.c
+++ b/deps/uv/src/unix/process.c
@@ -772,7 +772,11 @@ int uv__spawn_resolve_and_spawn(const uv_process_options_t* options,
     /* Try to spawn the new process file. If it fails with ENOENT, the
      * new process file is not in this PATH entry, continue with the next
      * PATH entry. */
-    err = posix_spawn(pid, b, actions, attrs, options->args, env);
+    int max_retries = 4;
+    int retries = 0;
+    do
+      err = posix_spawn(pid, b, actions, attrs, options->args, env);
+    while (err == EINTR && ++retries < max_retries);

test.parallel/test-child-process-spawnsync-validation-errors this is another interesting one, it was failing with EPERM whenever a test tried to set uid or gid for the child process. A minimal repro at https://gist.github.com/deepak1556/7b02c18eeede9cdcd5da55439e909e1b also validates this. Looking at the posix_spawn code, whenever posix_cred_info is filled by posix_spawnattr_set_uid_np, posix_spawnattr_set_gid_np, posix_spawnattr_set_groups_np it seems to trigger the following check

//https://github.com/apple/darwin-xnu/blob/8f02f2a044b9bb1ad951987ef5bab20ec9486310/bsd/kern/kern_exec.c#L3231

if (!kauth_cred_issuser(kauth_cred_get())) {
				error = EPERM;
				goto bad;
}

and kauth_cred_issuer seems to do the following check

// https://github.com/apple/darwin-xnu/blob/8f02f2a044b9bb1ad951987ef5bab20ec9486310/bsd/kern/kern_credential.c#L3500
/*
 * kauth_cred_issuser
 *
 * Description:	Fast replacement for issuser()
 *
 * Parameters:	cred				Credential to check for super
 *						user privileges
 *
 * Returns:	0				Not super user
 *		!0				Is super user
 *
 * Notes:	This function uses a magic number which is not a manifest
 *		constant; this is bad practice.
 */

Does this mean a super user needs to call the process triggering posix_spawn to set UID/GID bits ?

vtjnash · 2021-02-20T05:34:24Z

Yes, setuid / setgid should require being root, uid, or euid already, or fail with EPERM.

vtjnash · 2021-02-20T05:40:13Z

posix_spawn is not supposed to return EINTR, but if it was spawning multiple processes, we're probably getting many SIGCHLD notifications, so I wouldn't be surprised if we did get them here

vtjnash · 2021-02-20T05:54:41Z

Yeah, rr is a wish list feature, but still a bit hard to set up for CI, so while it would be great, it isn't something I'm fully prepared yet to champion. I just mention it because it makes (linux) debugging so much easier for particularly types of bugs, so I think that everyone should learn about it!

In libuv, there are still some frequent linux failures which keep it from being consistently green, though I'd fixed many of them this past year, so it is sometimes green. Unfortunately using rr also depends on a particular CPU feature, and many VM hosts don't enable that feature, including probably github actions (rr-debugger/rr#2402 (comment)). For configuration, yes, those would need to be temporarily made available somewhere for download. They're fairly compact, as they only need to include a copy of all files read and a bit of metadata.

deepak1556 · 2021-02-20T07:45:50Z

Yes, setuid / setgid should require being root, uid, or euid already, or fail with EPERM.

The test here https://github.com/nodejs/node/blob/master/test/parallel/test-child-process-spawnsync-validation-errors.js#L60 is trying to set uid of parent process, should that fail with EPERM ? It looks the check in kern_exec.c was only added in Catalina and higher which seem to require being root to make any uid changes even if the uid is same as the parent process...

but if it was spawning multiple processes, we're probably getting many SIGCHLD notifications, so I wouldn't be surprised if we did get them here

Yup that seems to be the case as the relevant test cases https://github.com/nodejs/node/blob/master/test/parallel/test-child-process-stdio.js in Node.js involve the PATH resolution stage https://github.com/descriptinc/libuv/blob/39968db6434f16f9fbd3c78f6c7392fd7ae9074e/src/unix/process.c#L693-L719 in the posix_spawn implementation.

jpcanepa · 2021-02-22T15:18:12Z

Unfortunate that jpcanepa is on vacation now, but I expect he'll dig into the findings soon after he's back. And better to get the release done. Sorry it delayed it so much without effect!

I'm back. I see @vtjnash has done a pretty good rundown already, but I'll try look on my end too.

vtjnash · 2021-07-26T18:00:14Z

For EINTR, I notice that is observed elsewhere also (https://reviews.llvm.org/D61096), though it potentially also should go away with #3251 (which blocks signals)

Refs: libuv#3107 (comment)

This kernel function is not permitted unless the process is setuid root, so disable this syscall. Falling back to fork/exec should be okay for the rare cases that the user decides they need to do setuid(getuid()) or setuid(geteuid()) for the child. Refs: libuv#3107 (comment)

Refs: libuv#3107 (comment)

This kernel function is not permitted unless the process is setuid root, so disable this syscall. Falling back to fork/exec should be okay for the rare cases that the user decides they need to do setuid(getuid()) or setuid(geteuid()) for the child. Refs: libuv#3107 (comment)

Refs: libuv#3107 (comment)

This kernel function is not permitted unless the process is setuid root, so disable this syscall. Falling back to fork/exec should be okay for the rare cases that the user decides they need to do setuid(getuid()) or setuid(geteuid()) for the child. Refs: libuv#3107 (comment)

Refs: libuv#3107 (comment)

This kernel function is not permitted unless the process is setuid root, so disable this syscall. Falling back to fork/exec should be okay for the rare cases that the user decides they need to do setuid(getuid()) or setuid(geteuid()) for the child. Refs: libuv#3107 (comment)

Fixes: #3050 Refs: #3086 Refs: #3064 Refs: #3107 Refs: #3064 This reverts commit 217fdf4, then fixes several issues with it: * remove error fast-cleanup code that triggers a nodejs bug Refs: #3107 (comment) * protect posix_spawn from EINTR This is not a documented valid error, but seems to have been observed. * ignore setuid/setgid syscall This kernel function is not permitted unless the process is setuid root, so disable this syscall. Falling back to fork/exec should be okay for the rare cases that the user decides they need to do setuid(getuid()) or setuid(geteuid()) for the child. Refs: #3107 (comment) * improve posix_spawn path search Ports the improvements in musl back to this function * fix some additional problems and formatting issues We previously might fail to start a watcher, in rare failure cases, resulting in a zombie that we would fail to kill. Also avoid creating the signal-pipe unless required (addresses a review comment from Apple) * fix fd->fd mapping reuse There was a chance that when duplicating the fd's into stdio_count+fd we might be closing a currently opened fd with that value.

several issues with it: * remove error fast-cleanup code that triggers a nodejs bug Refs: libuv#3107 (comment) * protect posix_spawn from EINTR This is not a documented valid error, but seems to have been observed. * ignore setuid/setgid syscall This kernel function is not permitted unless the process is setuid root, so disable this syscall. Falling back to fork/exec should be okay for the rare cases that the user decides they need to do setuid(getuid()) or setuid(geteuid()) for the child. Refs: libuv#3107 (comment) * improve posix_spawn path search Ports the improvements in musl back to this function * fix some additional problems and formatting issues We previously might fail to start a watcher, in rare failure cases, resulting in a zombie that we would fail to kill. Also avoid creating the signal-pipe unless required (addresses a review comment from Apple) * fix fd->fd mapping reuse

This reverts commit 39968db. This commit caused a number of failures in the Node.js test suite while attempting to release libuv 1.41.0. Refs: libuv#3086 Refs: libuv#3064 PR-URL: libuv#3107 Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl> Reviewed-By: Jameson Nash <vtjnash@gmail.com> Reviewed-By: Santiago Gimeno <santiago.gimeno@gmail.com>

Fixes: libuv#3050 Refs: libuv#3086 Refs: libuv#3064 Refs: libuv#3107 Refs: libuv#3064 This reverts commit 217fdf4, then fixes several issues with it: * remove error fast-cleanup code that triggers a nodejs bug Refs: libuv#3107 (comment) * protect posix_spawn from EINTR This is not a documented valid error, but seems to have been observed. * ignore setuid/setgid syscall This kernel function is not permitted unless the process is setuid root, so disable this syscall. Falling back to fork/exec should be okay for the rare cases that the user decides they need to do setuid(getuid()) or setuid(geteuid()) for the child. Refs: libuv#3107 (comment) * improve posix_spawn path search Ports the improvements in musl back to this function * fix some additional problems and formatting issues We previously might fail to start a watcher, in rare failure cases, resulting in a zombie that we would fail to kill. Also avoid creating the signal-pipe unless required (addresses a review comment from Apple) * fix fd->fd mapping reuse There was a chance that when duplicating the fd's into stdio_count+fd we might be closing a currently opened fd with that value.

This reverts commit 39968db. This commit caused a number of failures in the Node.js test suite while attempting to release libuv 1.41.0. Refs: libuv/libuv#3086 Refs: libuv/libuv#3064 PR-URL: libuv/libuv#3107 Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl> Reviewed-By: Jameson Nash <vtjnash@gmail.com> Reviewed-By: Santiago Gimeno <santiago.gimeno@gmail.com>

Fixes: libuv/libuv#3050 Refs: libuv/libuv#3086 Refs: libuv/libuv#3064 Refs: libuv/libuv#3107 Refs: libuv/libuv#3064 This reverts commit 217fdf4, then fixes several issues with it: * remove error fast-cleanup code that triggers a nodejs bug Refs: libuv/libuv#3107 (comment) * protect posix_spawn from EINTR This is not a documented valid error, but seems to have been observed. * ignore setuid/setgid syscall This kernel function is not permitted unless the process is setuid root, so disable this syscall. Falling back to fork/exec should be okay for the rare cases that the user decides they need to do setuid(getuid()) or setuid(geteuid()) for the child. Refs: libuv/libuv#3107 (comment) * improve posix_spawn path search Ports the improvements in musl back to this function * fix some additional problems and formatting issues We previously might fail to start a watcher, in rare failure cases, resulting in a zombie that we would fail to kill. Also avoid creating the signal-pipe unless required (addresses a review comment from Apple) * fix fd->fd mapping reuse There was a chance that when duplicating the fd's into stdio_count+fd we might be closing a currently opened fd with that value.

This reverts commit 39968db. This commit caused a number of failures in the Node.js test suite while attempting to release libuv 1.41.0. Refs: libuv/libuv#3086 Refs: libuv/libuv#3064 PR-URL: libuv/libuv#3107 Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl> Reviewed-By: Jameson Nash <vtjnash@gmail.com> Reviewed-By: Santiago Gimeno <santiago.gimeno@gmail.com>

Fixes: libuv/libuv#3050 Refs: libuv/libuv#3086 Refs: libuv/libuv#3064 Refs: libuv/libuv#3107 Refs: libuv/libuv#3064 This reverts commit 217fdf4, then fixes several issues with it: * remove error fast-cleanup code that triggers a nodejs bug Refs: libuv/libuv#3107 (comment) * protect posix_spawn from EINTR This is not a documented valid error, but seems to have been observed. * ignore setuid/setgid syscall This kernel function is not permitted unless the process is setuid root, so disable this syscall. Falling back to fork/exec should be okay for the rare cases that the user decides they need to do setuid(getuid()) or setuid(geteuid()) for the child. Refs: libuv/libuv#3107 (comment) * improve posix_spawn path search Ports the improvements in musl back to this function * fix some additional problems and formatting issues We previously might fail to start a watcher, in rare failure cases, resulting in a zombie that we would fail to kill. Also avoid creating the signal-pipe unless required (addresses a review comment from Apple) * fix fd->fd mapping reuse There was a chance that when duplicating the fd's into stdio_count+fd we might be closing a currently opened fd with that value.

cjihrig mentioned this pull request Feb 12, 2021

Release proposal: v1.41.0 #3086

Closed

bnoordhuis approved these changes Feb 12, 2021

View reviewed changes

vtjnash approved these changes Feb 12, 2021

View reviewed changes

santigimeno approved these changes Feb 12, 2021

View reviewed changes

cjihrig force-pushed the revert branch from 5775921 to 217fdf4 Compare February 12, 2021 22:04

cjihrig merged commit 217fdf4 into libuv:v1.x Feb 12, 2021

cjihrig deleted the revert branch February 12, 2021 22:05

VerteDinde mentioned this pull request Feb 18, 2021

revert: patch libuv to use posix_spawn on macOS (#27026) electron/electron#27797

Merged

5 tasks done

cjihrig mentioned this pull request Mar 30, 2021

uv_spawn is slow when parent process allocates mmap with MAP_JIT flag #3050

Closed

deepak1556 mentioned this pull request Jul 14, 2021

child_process spawn locks up renderer for 300 - 3000ms on signed app on macOS Big Sur electron/electron#26143

Closed

3 tasks done

vtjnash added a commit to vtjnash/libuv that referenced this pull request Jul 26, 2021

spawn: remove error fast-cleanup code that triggers a nodejs bug

6d7ca98

Refs: libuv#3107 (comment)

sergiou87 mentioned this pull request Sep 17, 2021

Upgrade to Electron v13 desktop/desktop#12978

Closed

sergiou87 mentioned this pull request Jan 7, 2022

spawn is slow on dev builds for macOS running on Apple silicon devices desktop/desktop#13639

Closed

vtjnash added a commit to vtjnash/libuv that referenced this pull request Jan 31, 2022

spawn: remove error fast-cleanup code that triggers a nodejs bug

6d6e7ce

Refs: libuv#3107 (comment)

vtjnash added a commit to vtjnash/libuv that referenced this pull request Feb 1, 2022

spawn: remove error fast-cleanup code that triggers a nodejs bug

c203584

Refs: libuv#3107 (comment)

vtjnash added a commit to vtjnash/libuv that referenced this pull request Feb 23, 2022

spawn: remove error fast-cleanup code that triggers a nodejs bug

22702ee

Refs: libuv#3107 (comment)

coderabbitai Bot mentioned this pull request Sep 12, 2026

feat(coding-agent): a run is done when its deliverable exists ID-Robots/clawbox#839

Merged

Conversation

cjihrig commented Feb 12, 2021

Uh oh!

bnoordhuis left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

vtjnash left a comment

Choose a reason for hiding this comment

Uh oh!

cjihrig commented Feb 12, 2021

Uh oh!

vtjnash commented Feb 12, 2021

Uh oh!

cjihrig commented Feb 12, 2021

Uh oh!

richardlau commented Feb 12, 2021

Uh oh!

deepak1556 commented Feb 18, 2021

Uh oh!

cjihrig commented Feb 18, 2021

Uh oh!

deepak1556 commented Feb 20, 2021

Uh oh!

vtjnash commented Feb 20, 2021

Uh oh!

vtjnash commented Feb 20, 2021

Uh oh!

vtjnash commented Feb 20, 2021

Uh oh!

deepak1556 commented Feb 20, 2021 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

jpcanepa commented Feb 22, 2021

Uh oh!

vtjnash commented Jul 26, 2021

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

7 participants

bnoordhuis left a comment •

edited

Loading

deepak1556 commented Feb 20, 2021 •

edited

Loading